Nvidia puts $5 billion into Ilya Sutskever's Safe Superintelligence, still with no public product

Nvidia will invest roughly $5 billion in Safe Superintelligence, the research lab Ilya Sutskever founded after leaving OpenAI, the companies said on July 27. In exchange, Nvidia gets what it calls "rare access into the company's closely guarded research" — a lab that has published no papers and shipped no product since its 2024 founding. SSI gets priority access to Nvidia's next-generation Vera Rubin platform, which Sutskever said would increase the company's available compute "by an order of magnitude" over the next 12 months. "We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so," Sutskever said. "We are confident that our big bet on the Vera Rubin platform will take us to the next level."

The deal also marks a hardware switch. SSI had built its research on Google's TPUs; the Nvidia investment moves it onto the GPU ecosystem most other frontier labs already use — the software tooling, networking, and supply chain Nvidia has spent a decade building around itself. For Nvidia, adding SSI to a customer list that already includes direct stakes tied to OpenAI's Ohio buildout and SK Group's Korea partnership extends a now-familiar pattern: invest in the lab, then supply the chips the investment is meant to be spent on.

What the $5 billion actually is

That last line is the genuinely strange part of this deal. A $5 billion check into a private company is, ordinarily, exactly the kind of event that resets a valuation — new capital in means a new price, unless the company is flush enough to dictate flat terms. SSI's valuation has moved once before: $1 billion raised at founding in September 2024 valued the company at $5 billion, and a $2 billion round in April 2025 repriced it sixfold to $32 billion. Reporting around this deal says that $32 billion figure stayed put through the Nvidia investment, which would mean Nvidia bought in at a price SSI hasn't grown into by any conventional measure over sixteen months. Nvidia was already a backer of that April round alongside Alphabet, Andreessen Horowitz, Lightspeed, and Sequoia — so this may be less a new round than an existing investor deepening a position at its own last price. Neither company has said so on the record, which is itself worth noting on a deal this size.

The company behind the number

That last note is a real disagreement, not a rounding error. Outlets covering this deal in the days after the announcement put SSI's headcount anywhere from about 33 to roughly 50, and a separate profile from mid-2025 had it near 20 — none of the figures traces to anything SSI itself has published, and the two most detailed accounts of the Nvidia deal (TechCrunch and DataCenterDynamics) don't state a number at all. Treat every headcount figure in this story, including this one, as an outside estimate rather than a confirmed fact.

What isn't in dispute: SSI has gone nearly two years without publishing a paper, shipping a product, or disclosing a dollar of revenue, and it is now valued at $32 billion on that record. The bet being made — by Nvidia and by SSI's existing venture backers — is almost entirely on Sutskever's pedigree as OpenAI's former chief scientist and co-founder, and on the theory that superintelligence research is worth funding as pure capability development, disconnected from the product cycle every other AI company is running on. That is a real, coherent bet. It is also one where the payoff, if there is one, remains entirely unverifiable from outside the company until SSI publishes something.

Against this newsroom's own register of closed AI funding raises, $5 billion sits comfortably above Moonshot AI's $3.5 billion Series F and CXMT's $8.6 billion Shanghai listing sits above it — a reminder that even a headline-grabbing strategic check from the industry's most valuable chipmaker is a mid-pack figure next to this year's biggest rounds and IPOs, not a record in its own right.

Why the pitch still works on investors

SSI's entire premise, since Sutskever's founding announcement in 2024, has been that safety and capability shouldn't be traded against a product roadmap — that the company would do one thing, "in a straight shot," with no intermediate product to distract from it. That framing is precisely what makes today's investment legible even without a product to point to: Sutskever spent nearly a decade as OpenAI's chief scientist before co-founding it, and the backers writing checks — Nvidia, Alphabet, a16z, Sequoia, Greenoaks — are underwriting his research judgment directly rather than a revenue forecast. It is a bet structurally closer to funding a scientist than funding a startup, which is unusual at this size and this valuation, and it is also exactly the kind of bet that is impossible to independently verify from the outside until the lab either publishes something or doesn't.

It also fits a wider pattern this newsroom has tracked all summer: chipmakers writing checks into the labs that then spend the money on their chips. AMD took its first-ever stake in a model lab this July, putting up to $5 billion into Anthropic as Anthropic committed to buying AMD's Instinct MI450 GPUs. Nvidia itself is already tied into OpenAI's Ohio data-center financing and SK Group's Korea buildout. The SSI deal is the same structure applied to a lab with no revenue at all to point to — which either means Nvidia's compute-for-equity playbook is now confident enough to run on pure research potential, or that the definition of an acceptable counterparty for this kind of deal has quietly gotten looser as the capital available to deploy has grown.